# import the datareader
import pandas_datareader.data as web
# import the date conversion function
import datetime

# initialize the start date
start = datetime.datetime(2014, 1, 1)
# initialize the end date
end = datetime.datetime(2016, 12, 30)
# get the Alibaba stock data for a given period from Yahoo Finance
data = web.DataReader("AMZN", 'yahoo',start,end)
# show the first 5 observations
data.head()

# save the data for future usage
data.to_csv("AMZN_data.csv")

# import the plotting library
import matplotlib.pyplot as plt
%matplotlib inline

# plot the opening and closing stock prices for 2015 only
plt.plot(data['Open']['2015'])
plt.plot(data['Close']['2015'])

# importing libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

from statsmodels.tsa.arima_model import ARIMA # ARIMA model
from statsmodels.tsa.stattools import acf, pacf #ACF and PACF

data.head()

# apply logarithmic transformation to the dataset
log_data = np.log(data)
# calculate the one-period difference
log_data_lagged = log_data - log_data.shift()
# drop the missing values
log_data_lagged.dropna(inplace=True)

plt.plot(log_data_lagged['Open']['2015'])
plt.plot(log_data_lagged['Close']['2015'])

# Plot ACF for Stock: Open with confidence treshholds
acf = acf(log_data_lagged['Open']['2015'])
plt.plot(acf)
plt.axhline(y=0,linestyle='--',color='gray')
plt.axhline(y=-1.96/np.sqrt(len(log_data_lagged['Open']['2015'])),linestyle='--',color='gray')
plt.axhline(y=1.96/np.sqrt(len(log_data_lagged['Open']['2015'])),linestyle='--',color='gray')
plt.title('Autocorrelation Function')
# take q = 1

# Plot PACF for Stock: Open with confidence bounds
log_pacf = pacf(log_data_lagged['Open']['2015'])
plt.plot(log_pacf)
plt.axhline(y=0,linestyle='--',color='gray')
plt.axhline(y=-1.96/np.sqrt(len(log_data_lagged['Open']['2015'])),linestyle='--',color='gray')
plt.axhline(y=1.96/np.sqrt(len(log_data_lagged['Open']['2015'])),linestyle='--',color='gray')
plt.title('Partial Autocorrelation Function')
# take p = 1

# ARIMA on Stock: Open
model = ARIMA(log_data['Open']['2015'], order=(1, 1, 1))
results_ARIMA = model.fit(disp=-1)
plt.plot(log_data_lagged['Open']['2015'])
plt.plot(results_ARIMA.fittedvalues, color='red')


from statsmodels.tsa.arima_model import ARIMA # ARIMA model
from statsmodels.tsa.stattools import acf, pacf #ACF and PACF

# Plot ACF for Stock: Close with confidence treshholds
acf2 = acf(log_data_lagged['Close']['2015'])
plt.plot(acf2)
plt.axhline(y=0,linestyle='--',color='gray')
plt.axhline(y=-1.96/np.sqrt(len(log_data_lagged['Close']['2015'])),linestyle='--',color='gray')
plt.axhline(y=1.96/np.sqrt(len(log_data_lagged['Close']['2015'])),linestyle='--',color='gray')
plt.title('Autocorrelation Function')
# take q = 2 in case of Stock: Close

# Plot PACF for Stock: Close with confidence bounds
log_pacf2 = pacf(log_data_lagged['Close']['2015'])
plt.plot(log_pacf2)
plt.axhline(y=0,linestyle='--',color='gray')
plt.axhline(y=-1.96/np.sqrt(len(log_data_lagged['Close']['2015'])),linestyle='--',color='gray')
plt.axhline(y=1.96/np.sqrt(len(log_data_lagged['Close']['2015'])),linestyle='--',color='gray')
plt.title('Partial Autocorrelation Function')
# take p = 2 in case of Stock: Close

# ARIMA on Stock: Close
model2 = ARIMA(log_data['Close']['2015'], order=(2, 1, 2))
results_ARIMA2 = model.fit(disp=-1)
plt.plot(log_data_lagged['Close']['2015'])
plt.plot(results_ARIMA2.fittedvalues, color='red')

plt.plot(log_data_lagged['Open']['2015'])
plt.plot(results_ARIMA.fittedvalues, color='red')
plt.plot(log_data_lagged['Close']['2015'])
plt.plot(results_ARIMA2.fittedvalues, color='red')

# save the fittedvalues to a new variables
predictions_log_lag = results_ARIMA.fittedvalues # stock: open
predictions_log_lag2 = results_ARIMA2.fittedvalues # stock: close
# To undo the differencing (shifting) we have to calculate the cumulative sum
predictions_almost_log = predictions_log_lag.cumsum()
predictions_almost_log2 = predictions_log_lag2.cumsum()

# check the type, to be sure it is a Series
type(predictions_almost_log)# series
type(predictions_almost_log2)# series

# Convert this series to DataFrame
predictions_almost_log = pd.Series.to_frame(predictions_almost_log)
predictions_almost_log2 = pd.Series.to_frame(predictions_almost_log2)
# check the type to make sure conversion was correct
type(predictions_almost_log)# frame now
type(predictions_almost_log2)# frame now

predictions_almost_log.columns=["OPEN"]
predictions_almost_log.head()

predictions_almost_log2.columns=["CLOSE"]
predictions_almost_log2.head()

# add that value to all rows
predictions_almost_log = predictions_almost_log + log_data['Open']['2015'].ix[0]
# first 5 observations
predictions_almost_log.head()

# add that value to all rows
predictions_almost_log2 = predictions_almost_log2 + log_data['Close']['2015'].ix[0]
# first 5 observations
predictions_almost_log2.head()

# Concatenate the very first observation and the above dataframe
predictions_log = pd.concat([log_data['Open']['2015'][0:1],predictions_almost_log])
predictions_log.head()

# Concatenate the very first observation and the above dataframe
predictions_log2 = pd.concat([log_data['Close']['2015'][0:1],predictions_almost_log2])
predictions_log2.head()

# transform back
predictions = np.exp(predictions_log)
predictions.head()

# transform back
predictions2 = np.exp(predictions_log2)
predictions2.head()

# plot real data
plt.plot(data['Open']['2015'])
# plot predictions in red circles
plt.plot(predictions,'ro')

# plot real data
plt.plot(data['Close']['2015'])
# plot predictions in green circles
plt.plot(predictions2,'go')

plt.plot(data['Open']['2015'])
plt.plot(predictions,'ro')
plt.plot(data['Close']['2015'])
plt.plot(predictions2,'go')